Does ChatGPT keep my file after I close the chat?

Yes. OpenAI's file FAQ says an uploaded file is kept for as long as the conversation it belongs to. Delete that chat and the file is removed within thirty days, unless a copy has been de-identified and disassociated from your account, or must be kept for security or legal reasons.

Does turning off training stop the upload?

No, and this is the point people miss most often. The training toggle governs whether your conversations improve future models. It does not change the upload itself: the file still travels to OpenAI, is still processed there, and is still stored with the chat until you delete it.

Are Temporary Chats safer for a sensitive file?

Somewhat. OpenAI says Temporary Chats are deleted after thirty days, are not used to train models, and leave no history or memories, though they may be reviewed to monitor abuse. The file is still uploaded and still processed on OpenAI's servers, so this reduces retention rather than removing exposure.

Do business plans handle uploads differently?

Yes. On Business, Enterprise, Edu, and the API, OpenAI says inputs and outputs are not used to train its models by default, and those plans offer additional administrative controls. The upload still happens and the file is still stored, so the difference is in how it is used, not whether it travels.

What is the safest way to ask about a private file?

Reduce what you send, or send nothing. Deleting columns you do not need before uploading genuinely lowers exposure. For a file that should not travel at all, use a tool that never receives it — Paperswift reads the export inside your browser and sends only your column names and their types.

Is it safe to upload a spreadsheet to ChatGPT?

Q: Can I upload customer data to ChatGPT?

That is a question about your agreements, not about ChatGPT's features. If a contract, policy, or regulation governs where that data may travel, uploading moves it somewhere new regardless of your settings. Check the obligation first; no privacy toggle inside the product can satisfy a commitment made outside it.

What actually happens when you upload a file?

ChatGPT cannot read a spreadsheet without receiving it. When you attach a CSV or Excel file, it is uploaded to OpenAI, and the analysis runs there — in a sandbox on OpenAI's infrastructure, not on your computer. That is not a policy choice; it is how the feature works.

Once the file arrives, it is stored with the conversation. OpenAI's file uploads FAQ says files are kept for as long as the chat they belong to, and that deleting the chat, the custom GPT, or your account removes the file within thirty days. Two exceptions are named: copies that have already been de-identified and disassociated from your account, and data OpenAI must keep for security or legal reasons.

There are practical ceilings too. A spreadsheet upload works up to roughly fifty megabytes, free accounts are limited to a few file uploads a day, and paid accounts have a rolling cap on uploads over each three-hour window. Those limits shape what you can do; they are not what decides whether you should.

Does ChatGPT train on the spreadsheet you upload?

On a personal plan, by default, it may. OpenAI's help pages say that on Free, Plus, and Pro accounts data sharing is enabled unless you turn it off, and the file FAQ confirms that the content used to improve models can include uploaded files.

The opt-out is real and takes about fifteen seconds. In the web app: your profile icon, then Settings, then Data Controls, then switch off "Improve the model for everyone." The Data Controls FAQ documents the same path on mobile, notes that the setting applies to your whole account across devices, and confirms you can change it at any time.

Business, Enterprise, Edu, and API usage sit the other way round: OpenAI says content from those offerings is not used to train its models by default. If your company already pays for one of those, the training question is largely settled for you.

Can I upload customer data to ChatGPT?

This is where the honest answer stops being about ChatGPT. The product's settings can tell you whether a file trains a model and how long it is retained. They cannot tell you whether you are allowed to send it in the first place.

If the spreadsheet holds customer records, payroll, patient information, or anything covered by a contract or policy about where data may live, that obligation was made outside the product and no toggle inside it can satisfy it. The question to answer first is whether moving this data to a new processor is permitted at all — and if you need to ask a colleague, ask before uploading rather than after.

For a great many files, none of this bites. Yesterday's marketing export, a public dataset, your own project's numbers: upload them and get on with the work.

How to reduce the exposure if you do upload

Four things genuinely help, in rough order of effect:

Send less. Delete the columns the question does not need before you upload. A file without the email addresses and account numbers is a smaller problem if anything ever goes wrong with it.

Use a Temporary Chat. OpenAI says these are deleted after thirty days, are not used for training, and leave no history or memories, though they may still be reviewed to monitor abuse.

Turn training off, if you are on a personal plan and have not already.

Delete the chat when you are done, which starts the thirty-day clock on removing the file with it.

What none of these do is un-send the file. They shrink how much you sent, how long it stays, and what it is used for — real improvements, all of them, but improvements on top of an upload that still happened.

When the honest answer is no

Some files should not be uploaded to any cloud tool, and no amount of configuration changes that. The tell is usually simple: if you would have to check a contract before attaching it, that is your answer.

For those files the options are to work without an AI tool at all, or to use one that never receives the file. Paperswift is the second kind. You open the CSV or Excel export in your browser, the query runs on your own machine, and the only thing sent when you ask a question is your column names and whether each column holds text, numbers, or dates — never the values in any row. There is no upload to configure, because there is no upload. The privacy page sets out exactly where that line falls.

That comes with real trade-offs. ChatGPT is far broader: it reasons about anything, writes and runs Python against your data, and folds the analysis into whatever you are writing next. Paperswift answers questions about a file and does nothing else. How it works covers the mechanics, and ChatGPT alternative for CSV & Excel compares the two directly. If you are still choosing between assistants rather than avoiding the upload, ChatGPT, Claude or Gemini for spreadsheet analysis is the more useful page.

The short version

Uploading a spreadsheet to ChatGPT is safe enough for most files, provided you know what you are agreeing to: the rows go to OpenAI, the file is stored with the chat, and on a personal plan it may improve future models unless you say otherwise. Turn training off, delete chats you no longer need, and send fewer columns than you think you need.

For the file where that is not good enough, the question was never which settings to use. It was whether the file had to leave your machine at all.